search

ChuhuaW / SGNet.pytorch

Pytorch Implementation for Stepwise Goal-Driven Networks for Trajectory Prediction (RA-L/ICRA2022)
116 stars 16 forks source link

Pre training model #9

Closed S1tephen closed 2 years ago

S1tephen commented 2 years ago

Hello, due to hardware reasons, I have been training for a long time. Could you share the pre training model

YuchenWang commented 2 years ago

We will release the checkpoints in the next update. Stay tuned!

S1tephen commented 2 years ago

Thank you. I'm looking forward to it

S1tephen commented 2 years ago

Hello, I encountered a bug. When I modify the variable（batch_size） of function build_data_loader(args, phase='train',batchsize=None), the dimension of the loaded dataset will be reduced by one dimension. Where should I modify batch szie. I look forward to your reply

ChuhuaW commented 2 years ago

You should not change the batch_size for ETH/UCY because we ordered the batches by the first_history_index (because certain sequences have enc_steps < 8). More information can be found in BiTrap. If you want to modify the batch size, you need write your own dataloader.